Model
Model explorer

Grok-1

OPEN
SpaceXAI · Grok 1 family · released Mar 17, 2024

xAI's first LLM; raw pretrained MoE base checkpoint (Oct 2023 data cutoff), open-sourced under Apache 2.0 on Mar 17 2024.

ReasoningCodingVisionFunction callingTool useAgentic
505.1
Elo · rank #302
Parameters
314B
Active params
86B (MoE)
Context
8K tokens
Architecture
Mixture-of-Experts Transformer (8 experts, 2 active per token, 64 layers, RoPE)
License
Apache License 2.0
Languages
API price (in/out)
No hosted API
Modalities
text
Benchmark results
Bar shows position within the tracked field; marker = field best
GSM8KMath62.9%#109
best: Llama 3.1 405B · 96.8%
HumanEvalCoding63.2%#94
best: Claude Opus 4.5 · 99.4%
MATH-500Math23.9%#174
best: GPT-5 · 99.4%
MMLUKnowledge73.0%#130
best: OpenAI o3 · 92.9%
Run it locally
VRAM @ Q4
180 GB
VRAM @ FP16
628 GB
Fits on (Q4)
M3 Ultra 512GB
Needs well over 100GB even at aggressive quantization — nowhere near a single RTX 4090.
Quantizations
GGUF
Grok 1 family
Elo progression across releases
API price weights · each benchmark row carries its own source badge (see methodology)